Some relevant concerns (which includes why it’s not possible)

• The privacy thing really does get in the way of the context management thing. Like in a big way. Not completely but it makes it an order of magnitude harder.
• It is very, very hard to prove on whatsapp that “we can’t read your messages”, because users have no guarantee that the server is the one actually sending responses. This could be a problem that is solved, but it would need to become its own business, which involves deep devops and security, which I could do but don’t have a lot of reps in.
  • If I use twilio (as an example), the following flow is possible
    • Have an open source endpoint that proves it’s directly receiving the message from twilio (since twilio sends a hash that includes the endpoint URL), and that deletes the message from all logs on twilio after it’s been received. It could also make any calls required to LLMs. This has the following problems:
      • As is currently the case in all backend functions, there is no way to prove that the open source code is actually running on the endpoint. This is petitionable and solvable.
      • More importantly and intractably - the endpoint requires history of all previous journal entries for it to work. There is no way to share this context with the server in any way that has attestable non-disclosure.
• Building on whatsapp would be much cleaner and easier! And solve so many other challenges.
• Privacy seems like an important feature - to the user